Fork-and-fill bootstrap for AI coding agents — MIT
AI coding agents break your rules.
This blocks the commit instead of asking nicely.
One command drops a .claude/ setup and a pre-commit gate into your repo.
No framework, no runtime, no registry to install from — what you get is plain files you own outright.
.env, or produced a commit with type errors still in it.CLAUDE.md, so results depend on whose session it was.Rules that live only in a document get ignored eventually. This scaffold turns them into a gate that blocks the commit, and plants it in the repo.
curl -fsSL https://raw.githubusercontent.com/leeyudok/agents-scaffold/main/bin/agents-scaffold.sh \
| bash -s -- --stack nextjs --yes
$ echo 'DB_PASSWORD=hunter2' > .env
$ git add -f .env app.py && git commit -m "feat: add config"
Blocked: a .env-type file is staged. Commit is not allowed.
This is not the same as writing “don't commit secrets” in a doc. .git/hooks/pre-commit is
what stops it, so it holds whichever AI tool you use, and it holds when a human commits by hand.
| Harness | Measured | baseline | full |
|---|---|---|---|
| Claude Code | 2.1.239 | holds | rules, skills, subagents, hooks |
| Codex | codex-cli 0.149.0 | holds | skills not discovered |
| Antigravity | agy 1.1.18 | holds | experimental |
baseline means the AGENTS.md policy body plus a real
.git/hooks/pre-commit gate and CI — it holds on every harness, and it holds when a
human commits from the terminal. full adds that harness's native layers. Support
status is tracked in a machine-readable
manifest
that CI fails when a claim goes 90 days without re-measurement.
.claude/memory/ and load next session.